David E. Culler University of California, Berkeley August 2, 2011

What can we put on the table? 2 8/2/2011CPS-PI-11

Sustainability?  “Sustainable development should meet the needs of the present without compromising the ability of future generations to meet their own needs”  Our Common Future, World Commission on Environment and Development, United Nations, /2/2011CPS-PI-11

Quantifying it - California Law  AB 32  Reduce GHG emissions to 1990 levels by 2020  Governor’s executive order S-3-05 (2005)  80% reduction below 1990 levels by 2050  Renewable Portfolio Standard  33% renewables by 2020, 20% biopower procurement  480 => 80 mmT CO2e in 40 years  Population expected to grow from 37 => 55 million  Economic growth 4 8/2/2011CPS-PI-11

5 8/2/2011CPS-PI-11

What it amounts to … 6 8/2/2011CPS-PI-11

Can we do this? 7 8/2/2011CPS-PI-11

But 8 8/2/2011CPS-PI-11

My Take …  We can’t build our way to sustainability  The discussion is focused on the Physical  New Windows, Materials, Motors, Biofuels, …  Need to make the best of what we have, or will have, and how we use it…  This takes observation, intelligence and control  Demonstrable CPS technology on the table  Efficiency & Supply-Following Loads  10 years to innovate, 30 years to scale 9 8/2/2011CPS-PI-11

What’s on the Table? – buildings 10 8/2/2011CPS-PI-11 Deployed Demonstrated Development Research Concept

4 Part GHG Reduction Plan  Efficiency 11 8/2/2011CPS-PI-11

4 Part GHG Reduction Plan  Efficiency  Electrify 12 8/2/2011CPS-PI-11

4 Part GHG Reduction Plan  Efficiency  Electrify  Decarbonize the electricity  Decarbonize the fuel 13 8/2/2011CPS-PI-11

4 Part GHG Reduction Plan  Efficiency  Electrify  Decarbonize the electricity  Decarbonize the fuel 14 8/2/2011CPS-PI-11

All required for even 60% reduction but still fall short of 80% 8/2/ CPS-PI-11

Efficiency 16 8/2/2011CPS-PI-11

Electrification 17 8/2/2011CPS-PI-11

Low Carbon Electricity Options 18 8/2/2011CPS-PI-11

Build Rate 19 8/2/2011CPS-PI-11

The Renewables are there 20 8/2/2011CPS-PI-11 43% agriculture, 3.4% urbanized

The Problem: Supply-Demand Match 21 Baseline + Dispatchable Tiers Oblivious Loads Transmission Generation Demand Distribution 8/2/2011CPS-PI-11

Load-following Supply (?) 8/2/2011CPS-PI Source: EEX spot prices. Growing proportion of renewables leads to higher price volatility. October 2008 to March 2010: >90 hours with negative prices; highest price reached: +€500/MWh, lowest -€500/MWh Daily minimum price (indicated in red when negative) Daily maximum price €/MWh

… and in CA 8/2/2011CPS-PI-11 23

Zero Emissions Load Balancing (ZELB)  Just the emissions from the natural gas to firm the 33% renewables exceeds GHG target  Even with 50% with natural gas & 50% with some yet-to-exist storage 24 8/2/2011CPS-PI-11

We Understand Supply 25 8/2/2011CPS-PI-11

Not demand 26 8/2/2011CPS-PI-11

Limits to Renewable Penetration  Variability, Intermittency of Supply  Visibility into Availability of Supply  Ability of Loads to Adapt  Algorithms and Techniques for Reactive Load Adaptation  Capability of the Infrastructure to maintain the match 8/2/2011CPS-PI-11 27

CPS and the 4 Part GHG Reduction Plan  Efficiency  Electrify  Decarbonize the electricity  Decarbonize the fuel 28 Monitoring, Analysis, Modeling, Waste Elimination, Power Proportionality, Optimal Control Intelligence, Communication, adaptation in Everything ZELB. Supply-Following Loads, Energy SLA, Cooperative Grid 8/2/2011CPS-PI-11

CPS Technology …  National-scale Physical Information Service  Data collection, streaming, archiving, querying  Modeling, Analysis, Control  Representation, metadata, life-cycle  Use others’, contribute yours 29 8/2/2011CPS-PI-11

Some starts 30 8/2/2011CPS-PI-11

sMAP: Uniform Access to Diverse Physical Information 8/2/2011CPS-PI Electrical Weather Geographical Water Environmental Structural Actuator Occupancy sMAP Modeling Visualization Continuous Commissioning Control Personal Feedback Debugging Storage Location Authentication Actuation Applications Physical Information REST API HTTP/TCP … JSON Objects

Internet Cell phone sMAP Gateway sMAP Modbus RS-485 sMAP sMAP Gateway EBHTTP / IPv6 / 6LowPAN Wireless Mesh Network sMAP Edge Router Temperature/PAR/TSR Vibration / Humidity AC plug meter Light switch Dent circuit meter Proxy Server EBHTTP Translation California ISO sMAP Gateway sMAP Resources Applications Google PowerMeter Weather sMAP Every Building Database sMAP ecosystem 8/2/2011CPS-PI-11 32

Cory CEC B2G Testbed 8/2/2011CPS-PI Whole Bldg MCL equip MCL infra MCL vac servers DOP HVAC Central vent office HVAC inst Lab 199 HVAC Plug loads Lighting Parking Lot

DOE MELS => Appliance Energy 8/2/2011CPS-PI LBNL Bldg of 1200 loads

DOE/UCB/Siemens Auto Demand Response  8/2/ Sutardja Dai Hall CPS-PI-11

BACNet => sMAP Sensors, Actuators Controllers Modem Siemens P2 over RS-485 Apogee Server Remote Login BACnet/IP Gateway sMAP Internet Sutardja Dai Hall 8/2/ CPS-PI-11

Data from 1 Modern Building  1358 control settings  Set points, Relays (lights, pumps, etc), Schedules  2291 meters/sensors  Power (building, floor, lights, chiller, pumps, etc)  Current, voltage, apparent, real, reactive, peak  Temp (rooms, chilled water, hot water)  Air volume  Alarms, Errors  2165 control outputs  Dampers, valves, min/max flow, fan speed, PID parameters  72 other 4+ million Commercial, 110+ million Residential 8/2/ CPS-PI-11

Current sMAP demography NameSensor TypePhysical LayerSense PointsChannels Cory Hall SubmeteringDent Three-PhaseModbus + Ethernet Cory Hall MeteringION and pQube MetersHTTP/Ethernet3150 Cory Lab TemperatureTelosB Ethernet48 Cory Lab MachinesACme Ethernet816 Cory Chilled WaterHeatXModbus + Ethernet111 Cory WeatherVaisala WXT520SDI-12 + Ethernet111 Soda Hall Sun BlackboxFan speed; environmentalHttp/Ethernet1084 Soda Lab MachinesACme Ethernet4080 Soda Lab PanelVeris E30Modbus + Ethernet142 Soda SCADA DataBarrington controlsRS-485/various“1”1670 LBNL Building 90Acme Ethernet Campus Power DataObvius Aquisuite; variousXML/HTTP/Ethernet4100 Citris SDH BACnetSiemens ApogeeBACnet/IP + RS-485“1”600+ Brower BuildinbgJohnsons ControlBACnet“1”1500 8/2/ CPS-PI-11

insert resample aggregate query streaming pipeline Page CacheLock Manager Key-Value Store Storage Alloc. Time-series Interface Bucketing RPC Compression readingdb SQL Storage mapper MySQL A Stream Engine Dent circuit meterDent circuit meter sMAP ~300k pts/sec 8/2/ CPS-PI-11

Scaling Out in the Cloud 8/2/2011CPS-PI Stream Management System Data Processing Data Source Data Sink Real Time Monitoring HBaseHive HDFS Map-Reduce Hadoop Interactive Workloads Analytics Workloads Columnar Data Storage Query Translator SQL  MR

15,000 pts and growing… 8/2/2011CPS-PI-11 41

CPS Technology  National-Scale Physical Info Service  Software foundations, platforms and solutions for Energy Efficiency and Agility 42 8/2/2011CPS-PI-11

Stages of Energy Effectiveness  Waste Not  Do Nothing Well !!!  Power Proportionality  Peak Performance : Power => Safety  Optimize Partial Load - from nothing to peakl  Sculpting  Identify the energy slack and utilize it  Negotiated Grid / Load / Human Interaction  Plan, Forecast, Negotiate, Manage 8/2/2011CPS-PI-11 43

8/2/ Lighting Servers PDUs, CRACs Our Buildings HVAC IT and Plug Load CPS-PI-11

8/2/ Power- Proportional Buildings ? 950 KW 1150 KW Min = 82% of Max Cory Hall: Office + Semiconductor + IT CPS-PI-11

8/2/ Power- Proportional Buildings ? 1.45 MW 2.02 MW Min = 72% of Max Stanley Hall: Office + BioScience - 13 NMRs CPS-PI-11

8/2/ Power- Proportional Buildings ? 62 KW 202 KW Min = 31% of Max LeConte Hall: Office CPS-PI-11

Re-Flash the HVAC … 48 Whole Bldg inst Lab 199 HVAC 8/2/2011CPS-PI-11 LoCal + ActionWebs

Learning-based Model Predictive Control  Mathematical model from Newton’s law of cooling  Model identified using semi-parametric regression dT/dt = -k r T - k c u(t) + k w w(t)) + q(t) change in temperature over time time constant of room AC cooling weather heating from occupants and equipment Temperature: Experimental (blue) Simulated (red) Heating from occupants and equipment 8/2/ (Aswani, Master, Taneja, Culler, Tomlin, Proc. of IEEE, 2011) CPS-PI-11

Living Lab…  LBMPC saved  30% energy when hotter outside  70% energy when cooler outside  Standard MPC was inconsistent  Two-position control was inefficient 50 Method Measured Energy Estimated Energy Tracking Error Temperature Variation Average External Load Two-Position Control Experiment LBMPC23.6kWh0.75°C0.13°C11.0°C Linear MPC30.5kWh0.62°C0.30°C11.0°C Two-Position32.6kWh35.1kWh0.61°C0.20°C11.0 °C LBMPC Experiment LBMPC11.8kWh13.3kWh0.86°C0.17°C7.2°C Linear MPC8.6kWh0.93°C0.21°C7.2°C Two-Position34.5kWh0.55°C0.19°C7.2°C 8/2/2011CPS-PI-11

LBMPC to Minimize Energy…  Estimates heating load using model  Best control that considers estimated occupancy (Aswani, Master, Taneja, Culler, Tomlin, 2011, submitted) 51 Cost AC Constraints Temperature Constraints Temperature Dynamics 8/2/2011CPS-PI-11

… or optimize to follow the supply 52 Generation Transmission Distribution Load VPG 8/2/2011CPS-PI-11

at Building-Scale, and beyond 53 Temperature Blue – Measured; Red - Simulated Heating from occupants, equipment, etc. 8/2/2011CPS-PI-11

Making Sense out of Sensors 54 °°°°°° °°°°°° °°°°°° 8/2/2011CPS-PI-11

Supply Air Fan VFD 55 8/2/2011CPS-PI-11

What’s really going on?  Bring comfortable air from outside  Cool it down till its really cold  Push it out everywhere thru VAVs that are at minimum opening  Reheat it to set point  So the empty rooms will be comfortable  This is going on everywhere!  And we supply perfect, precious energy to do it! 56 8/2/2011CPS-PI-11

And IT is no better … Core i7 Atom 333 Westmere 4/12/11 57 CPS 2011

Supply-Following Computational Loads 8/2/2011CPS-PI IPS Requests Power Background Processing (shiftable) Controllable Storage QoS (fidelity & latency) Availability Forecasts

Non-dispatchable, variable supply Power proportional, grid-aware loads NREL Western Wind and Solar Integration Study Dataset Pacheco wind farm Scientific computing cluster Energy-Availability Driven Scheduling 8/2/2011CPS-PI-11 59

Scheduling & Energy Storage 8/2/ CPS-PI-11

CPS Technology  National Physical Info Service  Software foundations for Energy Efficiency and Agility  Infrastructure for Energy Innovation 61 8/2/2011CPS-PI-11

Personalized Automated Lighting Control  Three controllable ballasts per fixture  ~5 zones per floor BACnet sMAP MySQL Python Django Python Control Process HTTP Gateway 8/2/ CPS-PI-11

Real Energy Savings 8/2/2011CPS-PI-11 63

From the agony of load-following … 64 8/2/2011CPS-PI-11

to Co-operative Energy Mgmt in a Cyber-Physical Grid 65 Load IPS Source IPS energy subnet Intelligent Power Switch Monitor, Model, Mitigate Deep instrumentation Power proportional Design Energy-Agile Control Shifting, Scheduling, Adaptation Forecasting Tracking Market Availability Pricing Planning 8/2/2011CPS-PI-11

Thanks 8/2/2011CPS-PI-11 66